Research Article

RNA interference is essential for cellular quiescence

See allHide authors and affiliations

Science  11 Nov 2016:
Vol. 354, Issue 6313, aah5651
DOI: 10.1126/science.aah5651

RNAi soothes the path to quiescence

Cells, such as those in stem cell niches and immunity memory cells, can exist in a nondividing, quiescent state, from which they can be aroused with the appropriate signal. RNA interference (RNAi) is an important epigenetic pathway in many organisms. Roche et al. found that in fission yeast, RNAi was an essential regulator of the quiescent state. RNAi promoted proper chromosome segregation during entry into quiescence. It also prevented inappropriate silencing of the ribosomal DNA during quiescence.

Science, this issue p. 10.1126/science.aah5651

Structured Abstract

INTRODUCTION

Quiescence is defined as the state of a nondividing (G0) cell that is still metabolically active and able to return to the cell cycle with full viability. Quiescent cells, from microbial resting structures to memory lymphocytes and stem cells, have predominant roles in the life cycles of many organisms. The ability of quiescent cells, given the appropriate environment or signals, to return to the cell cycle highlights their reversibility and plasticity. Despite their importance, the molecular mechanisms underlying quiescence entry, maintenance, and exit are not yet well understood.

RATIONALE

Quiescence can be viewed as a state of epigenetic plasticity. We therefore reasoned that specific epigenetic pathways may be involved in the control of the switch between quiescence and growth, as well as in the maintenance of the quiescent state. We investigated this proposition in the fission yeast Schizosaccharomyces pombe, a model organism for both epigenetics and quiescence. In this species, quiescence can be induced by a simple environmental signal, namely, nitrogen deprivation; wild-type cells retain viability for over 2 weeks and return to the cell cycle when put back in rich medium. S. pombe centromeres resemble those of higher eukaryotes, with heterochromatin at the pericentromeric regions marked by histone H3 lysine 9 (H3K9) methylation, and S. pombe possesses many conserved epigenetic pathways, such as RNA interference (RNAi).

RESULTS

In this study, we identify RNAi as a major requirement for quiescence. RNAi null mutants lose viability in both entry into and long-term maintenance of quiescence. In contrast, heterochromatin mutants do not show the maintenance defect, indicating that the mechanism by which RNAi regulates quiescence supersedes its known role in guiding heterochromatin formation at centromeres. To understand the molecular mechanism responsible for the G0 defect, we designed a genetic screen, based on iterative transitions between the cell cycle and quiescence, which allows the progressive enrichment and isolation of spontaneous suppressors of G0 defects. Parallelization of this screen can be used to obtain a large number of independent suppressors. We characterized 13 independent suppressors of G0 defects in cells lacking Dicer (dcr1Δ), which mapped to genes involved in chromosome segregation (class i), in heterochromatin formation [the silencing cryptic loci repression complex (CLRC) and Swi6HP1; (class ii)], and in RNA polymerase–associated factors (class iii). We found that class i suppressors rescue the mitotic defects caused by the loss of pericentromeric heterochromatin, and that a similar suppression can be observed by restoring this heterochromatin independently of RNAi. Proper segregation is important for mitosis at the transition between the cell cycle and quiescence. In contrast, class ii suppressors are especially important for quiescence maintenance. We found that during quiescence maintenance, dcr1Δ cells are defective in the release of RNA polymerase I (pol I). This leads to an accumulation of H3K9me at the ribosomal DNA (rDNA), which causes cell death, without the repeat recombination found in cycling cells. H3K9 methylation mutants or H3K9R histone substitution mutants bypass rDNA heterochromatinization and therefore restore the viability of RNAi mutants in G0. Class iii suppressors act upstream of this heterochromatinization by limiting the accumulation of RNA pol I at the rDNA.

CONCLUSION

We propose a model in which RNAi promotes RNA polymerase release in both cycling and quiescent cells: (i) RNA pol II release mediates heterochromatin formation at centromeres, allowing proper chromosome segregation during mitotic growth and G0 entry, and (ii) RNA pol I release prevents heterochromatin formation at rDNA during quiescence maintenance. The dual role of RNAi in the active cell cycle and in quiescence may therefore stem from a similar mechanism of RNA polymerase release, with a different consequence for heterochromatin formation depending on the genomic context and cellular state. Throughout eukaryotic evolution, RNAi and H3K9 methylation are always found together, indicating a codependency. Our model provides an explanation for this phenomenon, because the loss of RNAi would result in a strong selective pressure against H3K9 methylation in order to maintain rDNA epigenetic stability.

The dual role of RNA interference in the S and G0 (quiescence) phases of the cell cycle is based on RNA polymerase release.

The release of RNA pol II allows heterochromatin formation at centromeres, and the release of RNA pol I from rDNA avoids heterochromatin overaccumulation during quiescence maintenance. In the lower part of the figure, red and blue crooked lines represent small RNAs.

Abstract

Quiescent cells play a predominant role in most organisms. Here we identify RNA interference (RNAi) as a major requirement for quiescence (G0 phase of the cell cycle) in Schizosaccharomyces pombe. RNAi mutants lose viability at G0 entry and are unable to maintain long-term quiescence. We identified suppressors of G0 defects in cells lacking Dicer (dcr1Δ), which mapped to genes involved in chromosome segregation, RNA polymerase–associated factors, and heterochromatin formation. We propose a model in which RNAi promotes the release of RNA polymerase in cycling and quiescent cells: (i) RNA polymerase II release mediates heterochromatin formation at centromeres, allowing proper chromosome segregation during mitotic growth and G0 entry, and (ii) RNA polymerase I release prevents heterochromatin formation at ribosomal DNA during quiescence maintenance. Our model may account for the codependency of RNAi and histone H3 lysine 9 methylation throughout eukaryotic evolution.

In nature, most cells exist in a nondividing state (the G0 phase of the cell cycle). This state is commonly referred to as “quiescent” when cells are still metabolically active and able to reenter the cell cycle, given the appropriate signal. Examples of quiescent cells include those in stem cell niches (1), neuronal progenitor cells (2), and memory T cells (3). Yeasts and other microorganisms are also able to enter a reversible quiescent state in response to environmental cues (4), and in several pathogenic species, this state allows infection to persist in the host (5). Despite their importance, the molecular pathways involved in quiescence entry, maintenance, and exit are not well understood. Given that long-term stability and reversibility are hallmarks of quiescence, epigenetic mechanisms are likely involved, and we are studying their contribution to quiescence, using the fission yeast Schizosaccharomyces pombe as a model.

In S. pombe, synchronized entry into the reversible quiescent state is induced upon nitrogen starvation in the absence of a mating partner. Under these conditions, the rod-shaped cells divide twice without growth and differentiate into quiescent G0 cells, which retain viability and the ability to reenter the cell cycle for more than 1 month (4). G0 cells have a characteristic transcriptome (6, 7), morphology, and metabolic response (4, 8, 9). RNA interference (RNAi)–guided heterochromatic silencing is a major epigenetic pathway in S. pombe, which has one copy of each of the key enzymes involved: Dicer (Dcr1), Argonaute (Ago1), and RNA-dependent RNA polymerase (Rdp1). Transcription of pericentromeric repeats on both strands by RNA polymerase II is thought to generate partially double-stranded RNAs, which are cleaved by Dcr1 to small RNAs [21 to 23 nucleotides (nt)]. Small RNAs bound to Ago1 are targeted to homologous transcripts and lead to the recruitment of Rdp1 to generate more template double-stranded RNA. Subsequent engagement of the CLRC/Rik1 silencing complex, which includes the histone H3 lysine 9 (H3K9) methyltransferase Clr4 and the H3K4 demethylase Lid2, silences heterochromatic transcripts and reporter genes. This process is tightly regulated during the S phase of the cell cycle in order to ensure heterochromatin inheritance (10, 11) and is coupled to the DNA replication fork, which promotes heterochromatin spreading (12). RNAi mutants are viable but lose silencing of the pericentromeric repeats (13), resulting in chromosome segregation defects (14).

We induced prototrophic wild-type, dcr1Δ, ago1Δ, and rdp1Δ strains into G0 by nitrogen starvation and determined their viability at different time points up to 15 days (Materials and methods). Whereas the wild-type strain retained 88.5% viability after 15 days of G0, dcr1Δ viability was reduced to 53.3% at 24 hours and then gradually decreased to 7.9% at 15 days. rdp1Δ and ago1Δ cells had comparable, though less severe, viability defects (Fig. 1A and figs. S1 and S2). We distinguished two different stages in the loss of viability of RNAi mutant strains: (i) initial loss, which is characterized either by cells failing to enter G0 or by cells unable to exit G0, and (ii) subsequent gradual loss of viability over time, indicative of a defect in quiescence maintenance. In accordance with a major defect in G0 entry, there was an increase in the number of cells retaining a rodlike shape (Fig. 1, B and D), resulting in a lower number of cell divisions after nitrogen starvation (Fig. 1C). Missegregated DNA at the first division was readily detected by 4′,6-diamidino-2-phenylindole (DAPI) staining. Furthermore, whereas wild-type G0-exiting cells displayed a “schmoo” morphology at the first S phase (4), this was only observed for viable RNAi mutant cells (that form colonies when microisolated on rich medium) and not for the majority of inviable cells, supporting the idea that initial loss of viability reflects an inability to properly enter the quiescent state. Over time, refraction-negative (dead) and misshaped cells constituted an increasing proportion of RNAi-defective G0 cells (Fig. 1B). The dead cells were morphologically heterogeneous in comparison with, for example, tdp1Δ mutants in DNA repair (15), indicating that the loss of viability may be a complex effect.

Fig. 1 RNAi mutants lose viability in G0.

(A) Loss of viability at both G0 entry and during quiescence maintenance in prototroph dcr1Δ, rdp1Δ, and ago1Δ mutants [n = 5 biological replicates for wild-type cells (wt), 6 for dcr1Δ and rdp1Δ, and 7 for ago1Δ]. **P < 0.01 (t test) in comparisons between each mutant and wild-type cells for all time points, and between the time points 24hours (h) and 15 days. CFU, colony-forming units. (B) Microscopic observation of DAPI-stained RNAi mutants reveals an increased proportion of cells retaining a rod shape in early G0 (24 hours). After 15 days (d), whereas wild-type G0 cells look uniform, RNAi mutants display a variety of morphological defects and many dead refraction-negative and/or DAPI-negative cells. DIC, differential interference contrast. Upon G0 entry, RNAi mutants have (C) less of an increase in cell number relative to wild-type cells during the initial two divisions (24 hours G0; n = 2 biological replicates; **P < 0.01) and (D) a higher number of cells staying rod-shaped (24 hours G0; n = 2 biological replicates; *P < 0.05, **P < 0.01). All error bars indicate standard deviation.

To assess whether the G0 phenotype of the RNAi mutants was due to the loss of pericentromeric heterochromatin, we assayed swi6HP1Δ, chp2HP1Δ, and clr4Δ mutant strains for viability in G0. We observed that chp2Δ had a wild-type G0 phenotype, and that swi6Δ and clr4Δ strains were affected only during G0 entry and were subsequently viable (fig. S3, A to C). In parallel, we treated the wild-type strain with trichostatin A (TSA) for 24 hours before G0 induction; this treatment resulted in a loss of heterochromatin but only a minor loss of viability in G0 (fig. S3, D and E). These results indicate that whereas the loss of viability at G0 entry was common to RNAi and heterochromatin mutants, the loss of viability in quiescence maintenance in RNAi mutants was not due to loss of pericentromeric heterochromatin. The two distinct stages of viability loss therefore rely on two different mechanisms.

To further explore the function of RNAi during quiescence maintenance, we established a suppressor screen based on the sharp decrease in viability of dcr1Δ cells in G0, as follows. A population of dcr1Δ cells was induced into G0 by nitrogen starvation (EMM-N medium) for 1 to 3 days and was then put back into the cell cycle in rich medium (YES) for ~4 to 7 generations (Materials and methods). The population was then reinduced into G0, and the process was repeated, alternating between quiescence and growth for up to 20 cycles (Fig. 2A). In this procedure, cells carrying a spontaneous suppressor mutation have a fitness advantage at every cycle and are progressively enriched in the culture. We tested the protocol for proof of concept by using a known suppressor (Materials and methods). We subsequently performed the suppressor screen on >50 independent populations, and we identified 13 independent suppressors of dcr1Δ by whole-genome sequencing (16) and validation in backcrosses (Materials and methods; figs. S4 and S5). The suppressors fell into three main classes (Fig. 2B). Class i included mutants in chromosome segregation and kinetochore assembly: ndc80-R523L, ndc80-P431L, ndc80-R137N, and klp5-del. The Ndc80 mutants affect each of the functional domains of the protein: R137 is in the globular domain, P431 is in a Dis1-interaction site (17), and R523 is in the second long tail of unknown function. The klp5-del allele consists of a short deletion with a frameshift at codon 772, resulting in a different 772–814 C terminus. Class ii included mutants in the CLRC/Rik1 complex and HP1: rik1-K812*, raf2-G37V, swi6-W293*, swi6-T278K, rik1-T942K, clr4-R126*, and clr4-Y451* (where an asterisk indicates a stop codon). Class iii included mutants in RNA polymerase–associated factors: tbp1-D156Y and med31-ins. The insertion in med31-ins consists of the replacement of F23 by the sequence LVRIC, without frameshift. This phenylalanine is highly conserved from budding yeast Soh1pMED31 to human MED31p (18).

Fig. 2 Design of a G0 suppressor screen.

(A) Experimental design of the suppressor screen. A dcr1Δ population is alternated between cell cycling (in rich medium, YES) and quiescence (in nitrogen-deprived medium, EMM-N) every 1 to 3 days. Spontaneous suppressors, by their relative fitness advantage compared with the parental strain, become enriched during the alternations. After up to 20 alternations, individual clones are isolated and reassayed in G0 to check for suppression. (B) Summary of mutations present in 13 isolated suppressors of G0 defects in dcr1Δ, showing three classes of mutants: mutants in chromosome segregation (class i), in the CLRC/Rik1 complex and Swi6HP1 (class ii), and in RNA polymerase–associated factors (class iii). Each mutation in the list arose as an independent suppressor. Single-letter abbreviations for the amino acid residues are as follows: A, Ala; C, Cys; D, Asp; E, Glu; F, Phe; G, Gly; H, His; I, Ile; K, Lys; L, Leu; M, Met; N, Asn; P, Pro; Q, Gln; R, Arg; S, Ser; T, Thr; V, Val; W, Trp; and Y, Tyr. An asterisk indicates a stop codon.

Class i suppressors were mutants in the essential kinetochore Ndc80 subunit and the kinesin-8 ortholog Klp5, which promotes microtubule catastrophe (19). These mutants might be expected to suppress chromosome missegregation defects of RNAi mutants during the two accelerated divisions that S. pombe cells undergo at entry into quiescence (Fig. 1B), as well as in mitosis (14) and meiosis (20). Consistent with this hypothesis, we found that ndc80-R523L, ndc80-P431L, ndc80-R137N, and klp5-del are resistant to thiabendazole (TBZ, a microtubule poison), similarly to klp5Δ (19), and that these strains have fewer rod-shaped cells after 24 hours in G0 in a dcr1Δ background than the single-mutant dcr1Δ (fig. S6, A and B). Consistently, the lower suppression by ndc80-R137N may indicate that the Ndc80 loop has a more important role in the suppression, potentially via Dis1 by stabilizing the spindle-microtubule attachment (17). We were additionally able to phenocopy the missegregation phenotype in wild-type cells by treatment with TBZ during G0 entry (24 hours) (fig. S6E). Chromosome missegregation in RNAi mutants is caused by loss of pericentromeric heterochromatin, and we predicted that restoring pericentromeric heterochromatin in RNAi mutants (2123) should restore the G0-entry phenotype. Mutations in the H3K14-acetyltransferase Mst2 complex reduce pericentromeric transcription, allowing heterochromatin to be maintained in a RNAi-independent manner (21), and mst2Δ indeed restored the G0-entry phenotype of RNAi mutants (fig. S6F).

Class ii suppressors suppress the quiescence maintenance defect of dcr1Δ cells (fig. S5). All suppressors in the CLRC/Rik1 complex lose silencing of the pericentromeric dg/dh repeats and are therefore defective in heterochromatin formation; consistently, raf2-G37V is mutated in the RFTS (replication foci targeting sequence) (24), whereas swi6-T278K and swi6-W293* are mutated in the chromoshadow domain. We subsequently found that clr4Δ and swi6Δ suppressed dcr1Δ, ago1Δ, and rdp1Δ, as did rik1Δ (fig. S7), indicating that H3K9 methylation may be responsible for the progressive loss of viability during G0 in the absence of RNAi. To investigate this further, we performed chromatin immunoprecipitation followed by sequencing (ChIP-seq) of H3K9me2 in G0 wild-type and dcr1Δ cells (Fig. 3A and figs. S8 and S9). H3K9me2 localization in wild-type cells was prevalent at the centromere and subtelomeres, but not at the recently described transient heterochromatic islands found along chromosome arms (fig. S9), consistent with the observation that these islands disappear after nitrogen starvation (25). The dcr1Δ strain showed near-complete loss of centromeric H3K9me2 (fig. S8C), as well as loss of H3K9me2 at subtelomeric tlh loci (fig. S8B); however, a notable accumulation of H3K9me2 was observed at the ribosomal DNA (rDNA) locus (Fig. 3A and fig. S8A). Intriguingly, H3K9me2 accumulation matched the transcribed region of rDNA (Fig. 3A). H3K9me2 accumulation at the rDNA was also found in wild-type G0 cells at lower levels and depended on the H3K9 methyltransferase Clr4 (Fig. 3B). This H3K9me2 accumulation continued at longer G0 times in both wild-type and dcr1Δ cells, demonstrating that the difference between the genotypes increases as cells enter deeper into quiescence.

Fig. 3 Quiescence maintenance of dcr1Δ results in rDNA heterochromatinization by H3K9me.

(A) H3K9me2 ChIP-seq enrichment in dcr1Δ versus wild-type cells. In G0 cells, but not cycling cells, dcr1Δ heavily accumulates H3K9me2 at the rDNA locus [n = 2 biological replicates; cycling cell data from (30)]. (B) Validation by H3K9me2 ChIP-qPCR in wild-type and dcr1Δ cycling and G0 cells shows a stronger rDNA H3K9me2 increase in dcr1Δ G0 cells relative to that in wild-type G0 cells (n ≥ 3 biological replicates; **P < 0.01, t test). IP, immunoprecipitate. IN, input control. chr, chromosome.

To ascertain that the suppression effect of CLRC/Rik1 mutants is mediated by loss of H3K9me2 and not through a potential new target of the complex, we created a set of prototroph strains to assay histone H3 mutants in G0 (hht3Δ hht1-K9R hht2-K9R and partial H3K9R mutants hht3Δ hht1-K9R and hht3Δ hht2-K9R), keeping viability for over a week (Materials and methods; fig. S10A). The dcr1Δ hht3Δ hht1-K9R hht2-K9R strains restored G0 viability (fig. S10B), and the reduction of H3K9me2 in the dcr1Δhht3Δ hht1-K9R and dcr1Δ hht3Δ hht2-K9R strains alone was also sufficient to recover G0 viability (fig. S10, B and C). These results confirm that H3K9me2 is central to the quiescence maintenance defect of dcr1Δ cells. We then attempted to increase the levels of H3K9me at the rDNA to see whether the defect would be enhanced. Because the recruitment factor for Clr4 at rDNA in G0 is unknown, we decided instead to increase global Clr4 levels in G0. We replaced the promoter of clr4 at its endogenous location with the promoter of urg1 (fig. S11A), which is induced not only by uracil addition but also by nitrogen starvation (Materials and methods) (26). The resulting purg1-clr4 strain entered G0 with wild-type viability but overaccumulated H3K9me2 at rDNA and quickly lost viability, specifically in quiescence maintenance (fig. S11, B and C); in the dcr1Δ purg1-clr4 background, the H3K9me2 rDNA accumulation was increased even further, resulting in very low viability—almost complete inviability after a week in G0 (fig. S11, B and C). However, given that the accumulation of H3K9me2 is not specific to the rDNA in these strains, we cannot exclude the possibility that other genomic locations participated in the loss of viability in this assay by silencing essential genes.

Class iii suppressors also suppressed the quiescence maintenance defect (fig. S5). These suppressor mutations lay in RNA polymerase–associated factors: Tbp1TBP [in RNA polymerase (pol) I, II, and III)] (27) and the Mediator complex, a co-regulator of RNA pol II, which may also play a role in transcription by other RNA polymerases (28). These genes act upstream of rDNA heterochromatinization, because H3K9me2 at rDNA is reduced in dcr1Δ med31-ins and dcr1Δ tbp1-D156Y G0 cells (Fig. 4H). Notably, the strong suppressor med31-ins also suppressed G0 entry and TBZ sensitivity (fig. S5) and might therefore additionally act at centromeric heterochromatin, similar to mst2Δ and the Mediator subunit pmc2Δ (21). In support of this idea, distinct subunits of Mediator have been shown to play a role in silencing (29) and antisilencing (21). In dividing cells, Dicer is required for efficient RNA pol II termination at various genomic locations, including centromeres, transfer DNAs, and rDNA, and stalled RNA pol II accumulates in dcr1Δ at these loci (30) (fig. S12). We performed ChIP-seq using an antibody directed against RNA pol II [unphosphorylated C-terminal domain (CTD)] in wild-type and dcr1Δ G0 cells, but in contrast to the situation in cycling cells (30), no significant increase was detected at the rDNA in dcr1Δ G0 cells (fig. S12). ChIP–quantitative polymerase chain reaction (qPCR) using antibodies against the phosphorylated CTD pol II (S2P and S5P) also did not show significant differences. Small-RNA sequencing revealed that, in G0 cells, centromeric small RNAs were strongly reduced at pericentromeric regions (fig. S14, A and B), as were other small interfering RNAs, consistent with their production in the G1 and S phases of the cell cycle (31). Furthermore, in contrast to the situation in cycling cells (30), no small RNAs antisense to the rRNA were found (fig. S14C), and we did not detect any new loci of Dicer-dependent small RNA production in G0 cells (Materials and methods).

Fig. 4 dcr1Δ mutants are defective at releasing RNA pol I in quiescence maintenance.

(A) The main subunit of RNA pol I, Nuc1RPA190, was tagged with a (Gly)6 linker and 3×FLAG. (B) Probe location for ChIP-qPCRs. 1, rDNA promoter; 2, 5′ETS; 3, 18S; 4, 3′ETS. (C) RNA pol I enrichment at the rDNA was assayed by ChIP-qPCR using antibody (α) against FLAG in the nuc1-(Gly)6-FLAG background. This assay shows a higher occupancy of RNA pol I in wild-type cells than in dcr1Δ mutants among cycling cells, because of the reduced growth rate of dcr1Δ, and (D) a similar occupancy among early G0 cells (n = 2 biological replicates). (E and F) RNA pol I enrichment at 18S rDNA over the time spent in G0 (2 to 8 days) is reduced in wild-type cells, but dcr1Δ cells fail to release RNA pol I (n ≥ 2 biological replicates). prom, promoter. (G) The class iii suppressors, dcr1Δ tbp1-D156Y and dcr1Δmed31-ins, but not the class ii suppressor dcr1Δclr4Δ, show significantly decreased RNA pol I occupancy at rDNA in 8-day G0 cells (n = 2 biological replicates; *P < 0.05, **P < 0.01, t test). (H) The accumulation of H3K9me2 at rDNA repeats, assayed by ChIP-qPCR, is significantly reduced in class iii suppressors (n ≥ 3 biological replicates; **P < 0.01, t test).

We therefore decided to assay RNA polymerase I, which is responsible for rDNA transcription. RNA pol I occupancy at rDNA, assayed by ChIP-qPCR of tagged Nuc1RPA190 (Materials and methods; Fig. 4A), was lower in cycling dcr1Δ cells than in cycling wild-type cells, likely because of the slightly reduced growth rate of dcr1Δ cells resulting from the accumulation of DNA damage (10, 30) (Fig. 4C and fig. S15B). There was initially no significant difference in RNA pol I occupancy between the wild-type and dcr1Δ strains in G0, but we observed that RNA pol I occupancy was further reduced as wild-type cells entered into deeper quiescence, and that dcr1Δ cells did not experience this reduction, indicative of a failure to release RNA pol I (Fig. 4, D to F). This accumulation of RNA pol I was detected at the 5′ETS (external transcribed spacer), 18S, 5.8S, and 28S repeats, as well as at the very end of the 3′ETS next to the termination site Ter3. Notably, the accumulation of RNA pol I was not reduced in dcr1Δclr4Δ, suggesting that the failure to release RNA pol I lies upstream of H3K9me accumulation. In contrast, RNA pol I occupancy was strongly reduced in dcr1Δtbp1-D156Y and slightly reduced in dcr1Δmed31-ins, providing an explanation for their suppression effect, as well as for their reduced accumulation of H3K9me2 (Fig. 4, G and H).

The failure to release RNA pol I at rDNA in G0 may result in DNA damage, similar to the failure to release RNA pol II at centromeres and rDNA in the cell cycle (30). Because G0 cells do not rely on homologous recombination for DNA repair, rad22-YFP foci are not formed in quiescent 1c cells (8), although we observed that they strongly accumulated upon G0 exit in dcr1Δ cells (fig. S13, B and F). We therefore used H2A and H2A-S129phos (equivalent to γH2AX) antibodies to assay DNA damage directly in wild-type and dcr1Δ G0 cells by both ChIP-seq and ChIP-qPCR, and we observed an increase in rDNA in dcr1Δ cells (fig. S13, A and C). The H2A-S129phos/H2A ratio was reduced in suppressors that alleviate the RNA pol I defect, dcr1Δtbp1-D156Y and dcr1Δmed31-ins (fig. S13C); the lack of such reduction in dcr1Δclr4Δ suggests that DNA damage accumulation is correlated to stalled RNA pol I at the rDNA, but that the primary cause of death is not DNA damage itself but rather the accumulation of H3K9me2.

We further explored the distinction between RNA pol II release during the S phase and RNA pol I release during the G0 phase by constructing a dcr1Δreb1Δ double-mutant strain; Reb1TTFI is a factor that participates in RNA pol I termination and forces replication to occur in the same direction as transcription at the rDNA, thus avoiding collisions between DNA and RNA polymerase (32). Although reb1Δ cells are viable, the accumulation of RNA pol II in dcr1Δ cells may increase the need for Reb1 in order to avoid polymerase collisions and subsequent DNA damage at the rDNA. In accordance with this hypothesis, dcr1Δreb1Δ strains showed a strong negative epistatic interaction in dividing cells, resulting in extremely poor growth (fig. S15B); however, their viability was unaffected in G0 and was similar to that of the dcr1Δ single mutant (fig. S15A).

If RNA pol I is responsible for the quiescence maintenance defect of dcr1Δ cells, we reasoned that deletion of nonessential subunits of RNA pol I may destabilize it, and thus, similar to class iii suppressors, would allow RNAi-independent release. We therefore assayed rpa12Δ mutants in G0. Rpa12 is a nonessential subunit of RNA pol I that is required for polymerase termination (33); notably, we found that dcr1Δrpa12Δ suppressed the quiescence maintenance defect (fig. S15C), although it does not affect H3K9me2 levels at the rDNA. We were unfortunately unable to assay RNA pol I levels in these cells because the nuc1-FLAG rpa12Δ double mutants were inviable, most likely because RNA pol I becomes too unstable in this background. Thus, Rpa12 can be considered a class iv suppressor. Given that the suppressor screen is not saturated yet, we predict that sequencing many more of the >50 suppressors that we obtained will yield interesting alleles in RNA polymerase–associated factors in both class iii (general transcription) and iv (RNA pol I). We propose in particular that the specific recruitment factor responsible for the increase in H3K9me2 at rDNA in G0 could be a subunit of RNA pol I itself, providing a potential explanation as to why it would be difficult to recover such a suppressor.

We have identified RNAi as a previously unrecognized essential regulator of quiescence in S. pombe. RNAi both promotes heterochromatin formation at centromeres, allowing proper chromosome segregation during G0 entry, and prevents heterochromatin formation at the rDNA locus during quiescence maintenance (fig. S16). At G0 entry, missegregation results in cell death and can be suppressed by restoring heterochromatin or by strengthening segregation. In dividing cells, RNAi is required to release RNA pol II from rDNA and from pericentromeric heterochromatin, and in RNAi mutants, pol II must be removed from stalled replication forks by homologous recombination repair (10), resulting in a severe reduction in rDNA copy number (30). G0 cells possess a 1c DNA content and therefore use nonhomologous end-joining instead of homologous recombination for repair (8). Consistently, rDNA copies are not lost in RNAi mutants during quiescence. In quiescence maintenance, RNAi is required to release RNA pol I from rDNA. The failure to release RNA pol I results in over-recruitment of rDNA silencing factors, possibly by RNA pol I itself; in mammalian cells (34), rRNA interacts with SUV39H1 via NML (35), and RNA pol I interacts with the G9a methyltransferase (36). Stalled RNA pol I results in an overaccumulation of H3K9me2 at rDNA and in DNA damage, resulting in the loss of viability of these cells, similar to when essential genes are silenced by H3K9me2 (37). These defects can be suppressed by mutants in the silencing CLRC/Rik1 complex, the effector Swi6HP1, and specific mutants in RNA pol I nonessential subunits. Both RNA pol II and RNA pol I defects can be suppressed by specific alleles in the key transcription protein TBP (TATA-binding protein) and in the Mediator complex.

Thus, each class of suppressors illuminates important roles of RNAi in quiescent cells: in chromosome segregation, in heterochromatin formation and spreading, and in transcription. Because spontaneous mutations in essential genes are much more rare than in nonessential genes, we obtained relatively few class iii suppressors, and none from class iv. We expect that saturation of the G0 screen may uncover the precise components of RNA pol I, or rDNA–associated factors, involved in recruiting the CLRC/Rik1 complex for G0 rDNA silencing, as well as more “master regulator” genes that, like Dicer and TBP, regulate transcription by all RNA polymerases.

The rDNA locus plays a central role in aging (38), and H3K9me is a hallmark of rDNA silencing in plants (39) and mammals (35). The role of RNAi at rDNA may also be evolutionarily conserved; in mammalian cells, Dicer physically associates with rDNA (40), whereas in Candida albicans, Dicer is involved in precursor rRNA processing by cleaving the 3′ETS (41). In Neurospora crassa, quelling (RNAi) targets rDNA (42), and qiRNAs are generated at the rDNA locus by Dicer when cells are treated with alkylating agents or hydroxyurea (43). Furthermore, there is evidence that during evolution, RNAi and heterochromatin proteins (H3K9me and HP1) have been lost together, a loss that has happened independently in distinct fungal lineages such as budding yeasts of the subphylum Saccharomycotina (44), Ustilago maydis (Basidiomycota, subphylum Ustilagomycotina) (45), and the human pathogen Pneumocystis jirovecii (Ascomycota, subphylum Taphrinomycotina) (46). In budding yeasts, the loss of RNAi is correlated with the ability to acquire killer RNA viruses (47). Our results provide a possible explanation for a strong selective pressure to lose H3K9me-based heterochromatin upon RNAi loss, accounting for their codependency in eukaryotic evolution.

Our findings shed light on the anecdotal observation that some S. pombe mutants, such as dcr1Δ, do not keep well in long-term storage at low temperatures. We have found that repeated thawing-freezing cycles can select for spontaneous suppressors, which are widespread in laboratory strains of RNAi mutants. RNAi mutants also fail to differentiate into dormancy in asexual and sexual spores of Cryptococcus neoformans, an important human basidiomycete pathogen (48), and in the zygomycete Mucor circinelloides (49). In metazoans, such as Cænorhabditis elegans (50) and Drosophila melanogaster (51), Dicer mutants typically affect germ cells, which also spend long periods in quiescence. Given the importance of quiescence in the life cycle of unicellular and multicellular organisms, it is likely that epigenetic pathways will be found to be essential in neurons, germ cells, and cancer stem cells, which can spend many years in a quiescent state. Uncovering the mechanisms underlying the epigenetic regulation of quiescence thus opens a promising new field of study.

Materials and methods

S. pombe strain list and techniques

The list of strains used in this study is indicated in table S1. Crosses were performed by random spore analysis. De novo disruption of RNAi genes and C-terminal tagging of nuc1 were obtained by multiplex PCR, using primers indicated in table S2; the first round of PCR amplifies ~200 nt of homology regions (L1/L2, L3C/L4C and L5/L6 primers); 50 ng of each PCR product and 50 ng of pClonHph NotI-HF-digested plasmid were used in a second PCR step in order to amplify the hphMX cassette flanked by the homology regions. The PCR products were cleaned by column-purification on a QiaQuick PCR purification kit column (Qiagen), and 1 μg of DNA was used for transformation, without carrier DNA, using the Frozen-EZ kit (Zymo Research). Transformed yeast cells were plated on a nonselective YES (yeast extract with supplements, rich medium) plate overnight and the next-day replicated on selective YES medium containing 100 μg/ml hygromycin B (Sigma). All culture conditions were at 31°C.

For creating the clr4:NatMX-purg1-clr4 construct, we amplified the 5′ homology region with L1/L2 primers, the 3′ homology region containing the full clr4 ORF with V2-purg1-clr4/L6-clr4, and 800 nt of the urg1 promoter with the L5-purg1-800 and purg1-rev primers. 50 ng of each PCR product and 50 ng of pClonNat NotI-HF digested plasmid were used in the second multiplex PCR step to amplify the cassette with the L1/L6 primer pair, purified and transformed as above.

For making H3K9R mutants that would be usable in G0 assays, we decided to delete only one copy of H3, with the lowest expression according to the PomBase database, hht3, and both remaining copies were mutated and a selectable marker inserted either in 5′(N) (in-between the H3 hht gene and the H4 hhf gene) or in 3′(C) (following the 3′UTR of hht) (fig. S10A). The markers used were KanMX for hht1 and HphMX for hht2. In the first strategy, the mutation was introduced by amplifying the PCR products H3-hht(1/2)/H5M-hht(1/2) and H4M-hht(1/2)/H6-hht(1/2), and the products joined by PCR using the H3-hht(1/2)/H6-hht(1/2) primer pair. The 5′ homology region was amplified with the H1/H2 pair, and multiplex PCR, purification and transformation were performed as above to obtain the hht1-K9RN and hht2-K9RN strains. In the second strategy, the mutation was introduced by using the pairs ShortH3-hht(1/2)/H5M-hht(1/2) and H4M-hht(1/2)/H7-hht(1/2), which were joined by PCR using the ShortH3-hht(1/2)/H7-hht(1/2) pair. The 3′ homology region was amplified with the H8/H9 pair, and multiplex PCR, purification and transformation were performed as above to obtain the hht1-K9RC and hht2-K9RC strains.

All mutants were confirmed by Sanger sequencing of hht1 and hht2 loci: hht1-K9R replaces the 10th codon AAGLys with CGGArg, hht2-K9R replaces AAALys with CGAArg. The strains were crossed to the hht3Δ:NatMX strain, and the double mutants were then crossed together in order to obtain every combination of hht1-K9RN/Chht2-K9RN/Chht3Δ genotype, which were all viable with no growth defect and display wild-type level viability in G0 for more than 8 days; the viability was affected in several genotypes after 2 weeks of G0, therefore viability in fig. S10, B and C, was only assayed at the 24h and 8 days time-points. These triple-mutants were then crossed to dcr1Δ and every quadruple-mutant genotype was obtained. In parallel, the double-mutants hht1-K9RN/Chht3Δ and hht2-K9RN/Chht3Δ, in which only a subset of H3 in the cell should bear the K9R replacement, were also crossed to dcr1Δ and every resulting triple-mutant genotype obtained. All the crosses involving H3K9R mutations were performed by tetrad dissection to assay spore viability of each genotype and check for segregation distortion, using a Singer MSM400 micro-manipulator. All primers are given in table S2.

Tetrad analysis and spore viability

To assay spore viability, prototroph strains of identical genotypes (h− and h+) were crossed, and tetrads were micro-manipulated with a Singer MSM400 on a YES plate. Only spores from complete tetrads (4 spores) were isolated and then counted in the analysis.

In assays of co-lethality of dcr1Δ and rad51Δ (fig. S13, D and E), the dcr1Δ±sup strain was crossed to rad51Δ±sup. Only complete tetrads (4 spores) were micro-manipulated and dissected; furthermore, only the tetrads in which the genotype of all spores could be unambiguously determined were taken into account to calculate the viability of each genotype: wt±sup, dcr1Δ±sup, rad51Δ±sup, and dcr1Δrad51Δ±sup. The class iii suppressor med31-ins was not used in this assay, as it shows a reduced fertility causing tetrad dissection to become impractical; this phenotype is likely caused as it is a hypomorph mutation, considering that med31Δ is viable but completely sterile. Crosses followed by RSA readily recovered dcr1Δrad51Δmed31-ins triple-mutants, which like dcr1Δrad51Δtbp1-D156Y triple-mutants, are viable and rescue partially the strong growth defect of dcr1Δrad51Δ survivors.

G0 induction and survival assay

A cell culture of a heterothallic prototroph strain is cultured in EMM (Edinburgh minimal medium) then shifted to EMM-N (Edinburgh minimal medium without nitrogen) at 1-2 × 106 to induce G0-entry, similarly to previous methods (4, 6, 15). The temperature used in G0 was 31°C. At the indicated time points (1 day/24h, 8 days, 15 days), a 20 μl aliquot of the culture was placed as an inoculum on the side of a YES plate. Cells from the center of the inoculum were micro-manipulated with a Singer MSM400 to isolate ~100 single cells. The viability is then determined by the number of isolated single cells able to re-form a colony (colony-forming units), as has been described before (15). Physically isolating cells is important as different mutants can present distinct cell wall properties, resulting in “stickiness”, which would result in nonseparation if simply plated. In parallel, another aliquot was used to calculate the proportion of cells keeping a rod shape after 24h G0 using a hematocytometer, counting >4000 cells per experiment. All strains used for G0 assays were prototrophic.

We tested several staining agents and protocols (phloxin B, trypan blue, propidium iodide) as an additional means to estimate the viability of strains but none of these staining reagents yielded satisfactory results for G0 cells. This is likely explained by the changes in the cell wall and cell wall properties of quiescent cells, which can be seen ultrastructurally (4), but remain uninvestigated. While DAPI staining confirm the increase of dead DAPI-negative cells in a G0 culture losing viabilities, it is likely to overestimate viability as the presence of DNA in a cell is not an indication of viability.

The G0 phenotype of RNAi defective mutants was not likely due to a background mutation, as: (i) the prototroph progeny of a back-cross of RNAi mutants to a wild-type strain displayed the G0 phenotype in the RNAi mutant daughter strains, while wild-type daughter strains had a wild-type G0 viability (fig. S1, A and B); (ii) disrupting dcr1, ago1 or rdp1 de novo in the wild-type strain, with a different selection marker (hphMX), resulted in the same G0 phenotype, and actually slightly stronger in the case of ago1Δ (fig. S2); (iii) the catalytic-dead dcr1-5 mutant displayed a G0 phenotype identical to dcr1Δ (fig. S1C).

G0 suppressor screen

A population of dcr1Δ cells is alternated between rich medium (YES) and quiescence (in EMM-N) every 1-3 days. This has been done both on a population in liquid culture (10ml) and on plate (cell patch of 10μl). After 15-20 alternations, single cells are isolated from the population by plating on rich medium (YES) and assayed for G0 viability to determine if they harbor a suppressor.

The proof-of-concept of the suppressor screen was done as follows: in a liquid culture of dcr1Δ cells was introduced a subpopulation of dcr1Δmst2Δ cells at a ratio of 1:10000. After 15 alternations between growth and quiescence, >90% of isolated single clones from the total population were determined by PCR to harbor the dcr1Δmst2Δ genotype. Theoretically, if at every cycle 100% of dcr1Δmst2Δ cells survive vs. 55% of dcr1Δ cells, the enrichment should be (100/55)*exp(n) given n the number of cycles. With n = 15 we obtain a theoretical final concentration of dcr1Δmst2Δ cells of 78.4%; the higher observation is likely caused by an additional fitness advantage of dcr1Δmst2Δ cells during the growing cycles in rich medium.

We have observed that similarly, repeated freezing/thawing cycles or subculturing of RNAi mutants can select for spontaneous suppressors. We recommend to researchers wanting to investigate RNAi and G0 to start with fresh alleles, either by de novo disruption or by back-crossing. It is actually possible that some previous results in the literature using RNAi mutant strains may have been conducted with spontaneous suppressors. As most suppressors do not rescue pericentromeric silencing, they would not detected by the standard silencing assays but only when a prototroph is assayed for G0 viability.

TBZ experiments

In plating experiments the TBZ concentration used was 15 μg/ml. In G0 viability assays in the presence of TBZ, cells were G0-induced with EMM-N containing TBZ (5 μg/ml to 25 μg/ml) for 24h; after which the culture was washed twice in EMM-N before cells were assayed for viability as described before.

There is a correlation between the concentration of TBZ applied (0-25μM) and the loss of viability at G0-entry (Pearson r = -0.93). The viability of the cultures is unchanged after TBZ is removed for the rest of the experiment (fig. S6E).

Microscopy

Cells were dried on positively charged slides, stained with VectaShield HardSet medium with DAPI (Vector labs), and pictures were taken with an Axio Imager.M2 (Zeiss) microscope. For rad22-YFP strains, VectaShield HardSet medium without DAPI was used. G0-exit was induced by diluting an aliquot of EMM-N cells in rich YES medium (1:10), and the proportion of cells showing rad22-YFP foci, as well as the number of foci, were counted manually.

DNA-seq analysis

Genomic DNA from each suppressor was purified using the Genomic-tip 20/g column protocol (Qiagen) on 35-50 ml log-phase cells in YE medium. 1 μg of DNA was sheared on a Covaris S220 with a 350 bp target size for use in the TruSeq DNA PCR-free LT kit (Illumina). Barcoded libraries were pooled and sequenced on an Illumina MiSeq with a paired-end run type (150 bp read length).

FASTQ files were cleaned using SICKLE (version 1.210) (https://github.com/najoshi/sickle) for paired-end reads (pe –n parameters) and mapped to the S. pombe genome using BOWTIE2 (version 2.1.0) (default parameters) (http://bowtie-bio.sourceforge.net/bowtie2/index.shtml) (52). The SAMTOOLS suite (version 0.1.16) (http://samtools.sourceforge.net/) (53) was used to convert the alignment file to the BAM format, remove unmapped reads and PCR duplicates. The median coverage was calculated over the S. pombe genome in 1 kb bins by custom Perl and R scripts, and was determined to be >22X in each library (table S3).

SNPs were called using FREEBAYES (https://github.com/ekg/freebayes) (54) (parameters: -p 1 –standard-filters; version v9.9.2-10-g576bc70). The resulting VCF files were intersected with the VCF file obtained for the parental BR00 and BR15 strains to obtain SNPs specific to the suppressor strains, using the BEDTOOLS suite (version 2.17.0) (55). The VCFLIB suite (56) was then used to keep only SNPs with enough quality (QUAL > 10) and enough coverage (DP > 10), after which they were annotated based on the Pombase database. Each resulting SNP was then hand-checked to ignore low-confidence SNPs likely originating from sequencing errors occurring at long stretches of repeated nucleotides in intergenic regions and UTRs.

Copy number variations (CNVs) were checked for using custom Perl and R scripts by calculating regions with a coverage different from the median ± 3 interquartile ranges. Repeated regions at the subtelomeres, centromeres and rDNA are positive controls as the S. pombe assembly does not contain all repeat units and therefore reads acumulate on the units present in the assembly, resulting in a higher coverage. Every dcr1Δ suppressor also showed as expected a complete lack of coverage over the dcr1 ORF, confirming that they are indeed spontaneous suppressors and not wild-type contaminants (fig. S4). No other deletion/duplication was detected in the suppressors.

Validation of SNPs

All SNP-containing suppressed strains were back-crossed to a wild-type prototroph to isolate the SNP identified by DNA-seq. Those strains were further back-crossed to the initial dcr1Δ strain to confirm their causation of the suppression phenotype (fig. S5), as well as to a fresh dcr1Δ strain with a different antibiotic marker (hphMX) for the initial two SNPs obtained, ndc80-R523L and rik1-K812*.

To follow the SNPs in crosses, dCAPS-PCR primers were designed using the web-resource dCAPS Finder (http://helix.wustl.edu/dcaps/dcaps.html). A list of primers can be found in table S2. The ndc80-R523L strain additionally contained the SPBC119.15-R173P SNP, which was also back-crossed and observed to have no suppression effect on dcr1Δ.

ChIP

Chromatin immunoprecipitation was performed on cultures of 100 ml G0 (2 days) cells using the ZymoSpin ChIP kit (Zymo Research) with the following modifications to the protocol to adapt it to yeast cells: (i) the PBS-washed cross-linked cells were resuspended in 500 μl Chromatin Shearing buffer in a ZR Bashingbead 0.5 mm tube (Zymo Research) and lysed at 4°C by 4 cycles of vortexing for 4 min followed by a pause of 4min, after which the supernatant containing the chromatin was sheared using a Covaris S220 focused-ultrasonicator with a target size of 150-200 nt. (ii) the IP volume was reduced to 300 μl with 500 μl washes.

The cross-linking reaction was done at room temperature with shaking for 10 min for H3K9me2 and H2A, and for 30 min for RNA pol I and RNA pol II, followed by addition of fresh glycine at a final concentration of 0.125M for 10 min. The antibodies used were ab1220 (Abcam) for H3K9me2, ab817 (Abcam) for the unphosphorylated CTD of RNA pol II, ab5095 (Abcam) for RNA pol II S2P, ab5131 (Abcam) for RNA pol II S5P, ab13923 for H2A (Abcam), ab15083 for H2A-S129phos (Abcam) and M2-anti-FLAG (Sigma) for C-terminally tagged RNA polymerase I (nuc1-(Gly)6-3xFLAG-hphMX).

Primers for ChIP-qPCR are given in table S2. Libraries for ChIP-seq were constructed using the NEBNext ChIP-seq kit (New England Biolabs), barcoded, multiplexed, and sequenced on an Illumina MiSeq with a paired-end run type (100 bp read length). ChIP-seq data for cycling cells (H3K9me2 and RNA pol II S2P) was previously published (30) and the raw data were used in this study, and reanalyzed using the same pipeline as for the G0 data. FASTQ files were adapter-trimmed using CUTADAPT (version 1.6dev) (57), quality-filtered using SICKLE, then mapped to the S. pombe genome using BOWTIE2 (52). Unmapped reads and duplicate PCR reads were removed using SAMTOOLS (53). Genome coverage was computed using the BEDTOOLS suite (55), using both small-sized bins (20 bp) for Fig. 3 and figs. S8B, S8C, S9, S12, and S13 and larger bins (2 kb) for the genome-wide view in fig. S8A. Custom Perl and R scripts were used to calculate the IP enrichment over input after the addition of a pseudocount ψ = 1 RPM to each track to avoid wide variations in regions of low coverage. Our conclusions are not affected with ψ = 0 (no pseudocount addition), nor with higher values (ψ = 10). Also note that our conclusions are not affected when duplicate reads are kept. In Fig. 3 and fig. S12, the enrichment differential between dcr1Δ and wild-type cells is represented, and corresponds to the difference in the log2 enrichment of IP over input between these two strains (positive values indicating higher enrichment in dcr1Δ). In fig. S13, the H2A-S129phos/H2A ratio was used to calculate the enrichment differential between dcr1Δ and wild-type.

sRNA-seq

Total RNA was purified using the ZR Fungal/Bacterial MiniPrep kit (Zymo Research) and libraries constructed using the NEBNext Multiplex Small RNA Library Prep kit (New England Biolabs), barcoded, multiplexed and sequenced on an Illumina MiSeq (single-read 36 bp run-type). FASTQ files were adapter-trimmed and quality-filtered using the FASTX-toolkit (http://hannonlab.cshl.edu/fastx_toolkit/), mapped to the S. pombe genome using BOWTIE (58). Unmapped reads were removed using the SAMTOOLS suite (53), and genome coverage was computed using the BEDTOOLS suite (55). Custom Perl and R scripts were used for analysis and visualization of the data.

To search for potential new Dicer-dependent small RNA loci in G0, we compared for every annotation of the S. pombe the normalized read coverage in read per million (RPM) in wild-type and dcr1Δ G0 libraries looking for regions with (i) a minimum coverage RPM>5 in both wild-type library duplicates, and (ii) a coverage reduced by at least 66% in the dcr1Δ libraries. Only known loci passed this test (centromeric ncRNAs, centromeric boundary ncRNAs, tlh genes). Similarly, we repeated the analysis in an annotation-unbiased fashion by repeating the comparison for 100 nt windows over the whole genome with a minimum coverage RPM>2 in the wild-type and reduced at least 80% in dcr1Δ.

Custom scripts

All Perl and R custom scripts used to analyze and visualize the high-throughput sequencing data are available upon request.

Supplementary Materials

References and Notes

Acknowledgments: This work was funded by NIH grant R01 GM076396-08, the Howard Hughes Medical Institute and Gordon and Betty Moore Foundation Plant Biology Investigator Program (to R.M.), and the Agence Nationale de la Recherche grant ANR-13-BSV8-0018 (to B.A.). The authors acknowledge support from the Chaire Blaise Pascal (Fondation de l'École Normale Supérieure, France). We thank the Cold Spring Harbor Laboratory Woodbury Sequencing Facility. We thank everyone at the fission yeast database PomBase, because its resources have been extremely helpful for research in our laboratory. We also thank the reviewers for their helpful comments and suggestions. The raw data of the next-generation sequencing libraries reported in this paper are available in the Sequence Read Archive database under accession number SRP087488 (BioProject PRJNA341984). The authors report no conflicts of interest
View Abstract

Stay Connected to Science

Navigate This Article